Department of Electrical and Computer Systems …Wireless Local Area Network), ... CoS Class of Service CP Contention Period 2 MECSE-5-2004: ... NAV Network Allocation Vector

Department of Electricaland

Computer Systems Engineering

Technical ReportMECSE-5-2004

A Survey of IEEE 802.11 MAC Mechanisms for Quality ofService (QoS) in Wireless Local Area Networks (WLANs)

D. Pham, A. Sekercioglu and G. Egan

A Survey of IEEE 802.11 MAC Mechanisms for Qualityof Service (QoS) in Wireless Local Area Networks

(WLANs)

Minh-Duc Pham Y. Ahmet Sekercioglu Gregory K. Egan

March 29, 2004

Contents

1 Introduction 4

2 Legacy IEEE 802.11 MAC 52.1 Distributed Coordination Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.2 Point Coordination Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 IEEE 802.11e MAC Enhancements 73.1 Enhanced Distributed Coordination Function . . . . . . . . . . . . . . . . . . . . . . 83.2 Hybrid Coordination Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Distributed Approaches Based on DCF 94.1 Approaches based on priority . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4.1.1 Approaches based on IFS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104.1.2 Approaches based on CW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114.1.3 Approaches based on combination of IFS and CW . . . . . . . . . . . . . . . 124.1.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4.2 Approaches based on fair scheduling . . . . . . . . . . . . . . . . . . . . . . . . . . . 164.2.1 Matching priority to IFS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164.2.2 Matching priority to CW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174.2.3 Matching priority to BI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194.2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

5 Centralized Approaches Based on PCF 225.1 Priority-based approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225.2 TDMA-like approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235.3 Adaptive polling approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245.4 Contention-based multipolling approach . . . . . . . . . . . . . . . . . . . . . . . . 245.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

6 Summary and Conclusion 25

List of Figures

1 Two 802.11 superframes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 802.11 DCF access method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 A typical 802.11 superframe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 802.11e EDCF access method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1

MECSE-5-2004: "A Survey of IEEE 802.11 MAC Mechanisms for ...", D. Pham, A. Sekercioglu and G. Egan

5 802.11e EDCF queues vs 802.11 DCF queue per station . . . . . . . . . . . . . . . . 96 A typical 802.11e superframe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

List of Tables

1 Mapping from Priority to Access Category in EDCF . . . . . . . . . . . . . . . . . . 8

Abstract

Wireless local area network (WLAN) is becoming the edge network of choice in today’s net-work infrastructure. The IEEE 802.11, which involves the Medium Access Control (MAC) andphysical (PHY) layers, is so far the most widely used WLAN standard. However, it does notsupport Quality of Service (QoS) requirements of an increasing number of multimedia servicesbeing used on the networks. Therefore, a number of techniques for QoS support at MAC layerhave been proposed. This article reviews some of the presented methods and explains the prin-ciples, the advantages and drawbacks of each method. From this survey, it seems that selectingthe proper QoS mechanism and the best set of MAC parameters still remains a topic requiringmore research.

Keywords

IEEE 802.11, IEEE 802.11e, MAC (Medium Access Control), QoS (Quality of Service), WLAN(Wireless Local Area Network), DCF (Distributed Coordination Function), PCF (Point Coordina-tion Functions), EDCF (Enhanced Distributed Coordination Function), HCF (Hybrid Coordina-tion Function).

List of Abbreviations

AC Access CategoryACK AcknowledgmentADB Age Dependent BackoffAEDCF Adaptive Enhanced DCFAI Association InformationAID Association IDAIFS Arbitration Inter-Frame SpaceAIFSD AIFS DurationAP Access PointAR Association RequestARME Assured Rate MAC ExtensionBC Backoff CounterBCUT Backoff Counter Update TimeBEB Binary Exponential BackoffBI Backoff IntervalBSS Basic Service SetCA Collision AvoidanceCAP Controlled Access PeriodCF Contention FreeCFACK Contention Free AcknowledgmentCFB Contention Free BurstCFP Contention Free PeriodCFPRI Contention Free Period Repetition IntervalCoS Class of ServiceCP Contention Period

2


CSMA Carrier Sense Multiple AccessCTS Clear To SendCW Contention WindowCWmin Contention Window MinimumCWmax Contention Window MaximumDBRP Distributed Bandwidth Reservation ProtocolDC Deficit CounterDCF Distributed Coordination FunctionDCF/SC DCF/Shortened Contention WindowDDRR Distributed Deficit Round RobinDFS Distributed Fair SchedulingDIFS DCF Inter-Frame SpaceDLC Data Link ControlDRR Deficit Round RobinDS Distribution SystemDU Data UnitDWFQ Distributed Weighted Fair QueuingEDCF Enhanced DCFEIED Exponential Increase Exponential DecreaseFIFO First In First OutFTP File Transfer ProtocolHAD Hybrid Activity DetectionHCF Hybrid Coordination FunctionHIPERLAN High Performance LANIEEE Institute of Electrical and Electronics EngineersIETF Internet Engineering Task ForceIFS Inter-Frame SpaceIP Internet ProtocolISM Industrial, Scientific and MedicalISO International Organization for StandardizationJW Jamming WindowLAN Local Area NetworkLT Life TimeMAC Medium Access ControlM-DCF Modified DCFMF Multiplicator FactorM-PCF Modified PCFMPDU MAC Protocol Data UnitMPEG2 Motion Picture Experts Group 2MPX Motion Picture Extreme Compressed MoviesMS Mobile StationMSDU MAC Service Data UnitNAV Network Allocation VectorNRTDU Non-Real-Time Data UnitNTS Nothing To SendOFDM Orthogonal Frequency Division MultiplexingPC Point CoordinatorPCF Point Coordination FunctionPF Persistent/Persistence/Priority FactorPHY Physical layerPI Polling InformationPIFS PCF Inter-Frame SpaceP-MAC Priority-based MACPVT Packet Virtual Time

3


QoS Quality of ServiceQSTA QoS StationRCG Reservation Cycle GeneratorRF Reservation FrameRR Round RobinRTDU Real-Time Data UnitRTS Request To SendSAD Statistical Activity DetectionSCFQ Self-Clocked Fair QueuingSD Slow DecreaseSID Sequence IDS-PCF Slotted PCFSS Slave SchedulerSTA StationSVT Station Virtual TimeTBTT Target Beacon Transmission TimeTC Traffic CategoryTCMA Tiered Contention Multiple AccessTCP Transmission Control ProtocolTDMA Time Division Multiple AccessTOS Type of ServiceTXOP Transmission OpportunityUAT Urgency Arbitration TimeUDP User Datagram ProtocolUNII Unlicensed National Information InfrastructureUP User PriorityVDCF Virtual DCFWLAN Wireless Local Area Network

1 Introduction

Over the past few years, many multimedia applications such as voice telephony, streaming au-dio and video on demand have become increasingly popular under the Internet Protocol (IP)networks. However, IP was not designed to support multimedia services with stringent require-ments on minimum data rate, delay and jitter. The desire to use these multimedia applicationsover IP networks has led to the need for enhancing the existing networks with end-to-end QoSsupport. QoS is the ability to offer some persistent data transmission over the network withdifferent treatment for different traffic classes [36]. The Internet Engineering Task Force (IETF)is currently working on service differentiation at the IP layer to support various traffic classes.However, for optimal result, there is a need for QoS support from lower layers [52], especiallyData Link Control (DLC) layer. The IEEE 802.11 WLAN standard, which covers the MAC sub-layer of the data link layer and the physical layer, is gaining growing popularity, acceptance andis being deployed everywhere, such as hot spots in coffee shops, hotels and airports. However,the 802.11 standard does not currently provide QoS support for multimedia applications. As aresult, numerous proposals for QoS support at the MAC layer have been proposed by an activeresearch community. In this article, we present a survey of research efforts focusing on IEEE802.11 MAC QoS mechanisms for WLANs. The article is organized as follows: The legacy IEEE802.11 MAC is described in Section 2, while its 802.11e enhancements are presented in Section 3.Section 4 reviews the distributed mechanisms for QoS support based on Distributed Coordina-tion Function (DCF). The centralized mechanisms for QoS support based on Point CoordinationFunction (PCF) are evaluated in Section 5. The article concludes with a summary and conclusionin Section 6.

4


2 Legacy IEEE 802.11 MAC

An IEEE 802.11 WLAN can operate in two modes: ad-hoc or infrastructure. In ad-hoc mode,mobile stations (MSs) can directly communicate with each other. In the infrastructure mode, anaccess point (AP) allows the MSs to communicate with each other, and also connects them to adistribution system (DS). In IEEE parlance, the cluster of communicating MSs (and the AP) iscalled Basic Service Set (BSS). In the infrastructure mode APs, through the DS, link the BSSs to-gether. The traditional IEEE 802.11 MAC [1] contains algorithms to arbitrate media access (calledcoordination functions), which control the operation of MSs within a BSS to decide when a MSis allowed to send or receive frames via the wireless medium. Two coordination functions aredefined, which include the compulsory DCF based on Carrier Sense Multiple Access/CollisionAvoidance (CSMA/CA) and the optional PCF based on polling. The DCF may be used in bothad-hoc or infrastructure networks, while the PCF can only be used in the infrastructure networks.The DCF and PCF modes of operation are multiplexed in a superframe, which is repeated overtime. In a superframe, a Contention Free Period (CFP), in which PCF is used for medium ac-cess, is followed by a Contention Period (CP), in which DCF is used. A diagram of two 802.11superframes is shown in Figure 1. All the parameters such as Short Inter-Frame Space (SIFS),PCF Inter-Frame Space (PIFS), DCF Inter-Frame Space (DIFS), SlotTime, Contention WindowMinimum (CWmin), Contention Window Maximum (CWmax) depend on the selection of the un-derlaying physical layer. The IEEE 802.11 standard consists of a family of standards. The original802.11 standard [1] provides data rates up to 2 Mb/s at 2.4 GHz industrial, scientific, and med-ical (ISM) band. Its enhanced version IEEE 802.11b [3] achieves the data rates up to 11 Mb/s inthe ISM band. Another version, IEEE 802.11a [2] extends the data rates up to 54 Mb/s, usingorthogonal frequency division multiplexing (OFDM) at 5 GHz unlicensed national informationinfrastructure (UNII) band. However, regardless of the selected physical layer, following equali-ties are always valid: PIFS � SIFS � SlotTime and DIFS � PIFS � SlotTime. The operation ofDCF and PCF are described in details in Section 2.1 and Section 2.2, respectively.

CFP repetition period

Foreshortened CFP

Delay (due to busy medium �

DCFPCFB B PCF

DCF

Access Point broadcasts the Beacon Frame

NAV NAV

Busy Medium

B = Beacon Frame

Variable Length(per SuperFrame)

Access Point polls the stations

In this period, stations defer accessing the medium

Due to busy medium in the previous CFP repetition period

Contention Period

Content ion Fre e Period

Contention Period

Figure 1: Two 802.11 superframes

5


DIFS

DIFS

PIFS

SIFSBusy

Medium

Defer AccessSlot Time

Select Slot and decrement backoffas long as medium stays idle

Immediate access whenmedium is idle >= DIFS

Backoff Window Next Frame

Contention Window

Figure 2: 802.11 DCF access method

2.1 Distributed Coordination Function

The access method of DCF is shown in Figure 2. Each MS has a First In First Out (FIFO) trans-mission queue. A station can sense whether the medium is busy (a station is transmitting) or idle(there is no transmission). When a frame (or in ISO terms, a MAC Service Data Unit (MSDU),which is the unit of data entering the MAC layer from higher layer) reaches the front of a sta-tion’s transmission queue, the station checks the state of the medium. If the medium is busy, thestation waits until the medium is idle. After that, it waits for a duration of DIFS. If the medium isstill idle during the DIFS interval, the station initiates the backoff procedure. This is the CollisionAvoidance (CA) mechanism to reduce the probability of frame collisions if two or more stationsperceive the medium idle at the same time. Backoff Counter (BC) is chosen as a random integerin the uniformly distributed interval [0, CW], where CW is the current contention window of thisstation. The initial value of CW is CWmin. The Backoff Interval (BI) is the backoff time of thestation, which is equal the BC multiplied by the SlotTime. If the medium is idle for SlotTime, theBC is decremented by 1. When BC is 0, the station transmits the frame. If during the backoff pro-cedure, the medium becomes busy (another station finishes its backoff procedure and transmits aframe), the BC is suspended. This station will have to wait for the medium to become idle again,wait for extra DIFS duration before continuing the backoff procedure with the suspended BCvalue. If the transmitted frame is received successfully, the receiving station waits for a durationof SIFS and sends back an Acknowledgment (ACK) frame. The sending station, upon receivingthis ACK frame, resets its CW to CWmin, defers for DIFS and initiates a post backoff procedure,even if there is no frame waiting in the transmission queue. The post backoff procedure guaran-tees that there is at least one BI between transmission of two successive MSDUs. If the sendingstation does not receive the ACK frame after some time (either due to two or more stations fin-ish backoff process at the same time and the transmitted frames collide, the transmitted frameis lost or the ACK frame is lost), it assumes that the frame transmission is unsuccessful. It in-creases the CW to the new value of 2 � � CW � 1 �� 1, with upper limit of CWmax, waits for themedium to become idle again, waits for DIFS and performs backoff process with this new valueof CW. In case when an MSDU reaches the front of a station’s transmission queue and the stationis performing DIFS deferral or post backoff process, the frame is transmitted when the backoffprocedure completes successfully. If the MSDU reaches the front of the queue and the backoffprocedure has finished and the medium has been idle for at least DIFS, the frame is transmittedinstantly. The MSDU can be of any size up to 2304 bytes. A large MSDU can be fragmentedinto smaller frames. To solve the hidden station problem, an optional Request To Send/Clear ToSend (RTS/CTS) mechanism is introduced. Instead of transmitting a DATA frame after gainingaccess to the medium, a station sends a short RTS frame. The receiving station waits for SIFS andreplies with a short CTS frame. Upon receiving the CTS frame, the sending station waits for SIFSand transmits the DATA frame. The RTS and CTS frames stores information about transmissionduration of the DATA frame and are received by adjacent stations, which could be hidden fromsending and/or receiving stations. These adjacent stations set their Network Allocation Vector

6


(NAV) and restrain from transmitting for this duration, thus avoid collision.

2.2 Point Coordination Function

The PCF is only applicable for infrastructure WLAN, which consists of an AP and a number ofstations associated with it. The operation of stations is controlled by a station called Point Coor-dinator (PC). At the start of a superframe, the PC, which is usually located at the AP, generates abeacon frame periodically, whether PCF is active or not. Other stations know the time when thenext beacon frame will arrive, this time is designated as Target Beacon Transmission Time (TBTT).The beacon frame transports the management information to other stations to synchronize localtimers in stations and provide protocol related parameters. Stations can use the information inthe beacon frame to associate with the PC during CP to have their transmissions scheduled inthe next CFP. After the beacon frame, the PC waits for SIFS and polls the stations registered inits polling list by sending a CF-Poll frame to a station. If the PC has data or acknowledgmentdestined for this station, it piggybacks the CF-Poll frame on the DATA or CF-Ack frames, re-spectively. When a station receives the CF-Poll frame, it waits for SIFS and replies with a DATA,CF-Ack, DATA + CF-Ack or NULL frame (the NULL frame is transmitted when there is no datato send and no pending acknowledgment). If the polled station does not reply after PIFS, the PCpolls the next station. This is repeated until the CFP expires. The PC sends a CF-End frame toindicate the end of the CFP. A typical 802.11 superframe is shown in Figure 3.

NAV

D1+pollBeacon

SIFS

U1+ack

D 2+a c k+poll

U 2+a ck

D 3+a ck+poll

D 4+poll

U4+ackCF- E nd

Dx = Frames sent by PointCoordinator

Ux = Frames sent by polled

stations

PIFS SIFS SIFS SIFS

SIFS SIFS SIFSPIFS

No responseto CF- Poll Reset NAV

Contention Free Period Contention Period

Contention Free Repetion Period

Contention Free Maximum duration

Figure 3: A typical 802.11 superframe

3 IEEE 802.11e MAC Enhancements

To support QoS, the IEEE 802.11 Task Group E propose enhancements to the above standardin the IEEE 802.11e draft [4]. Within 802.11e, the Enhanced Distributed Coordination Function(EDCF) is an enhancement of the DCF in the legacy 802.11 while a new medium access mecha-nism called Hybrid Coordination Function (HCF) is introduced. The EDCF is part of the HCF.The HCF combines aspects of both DCF and PCF with the multiplex of CFP and CP in a 802.11esuperframe, which is repeated over time. EDCF is used only in CP, while HCF may be used inboth CP and CFP. The EDCF and HCF are described in Section 3.1 and Section 3.2, respectively.

7


3.1 Enhanced Distributed Coordination Function

Support for QoS is provided by the introduction of Traffic Categories (TCs). Each station has upto four Access Categories (ACs) to support up to eight User Priorities (UPs). One or more UPsare assigned to one AC. The mapping of UPs to ACs is shown in Table 1.

Priority Access Category Designation1 0 Best-effort2 0 Best-effort0 0 Best-effort3 1 Video probe4 2 Video5 2 Video6 3 Voice7 3 Voice

Table 1: Mapping from Priority to Access Category in EDCF

In Table 1, the priorities are sorted in ascending order with the exception of relative priority 0being placed between priorities 2 and 3. This is originated from IEEE 802.11d bridge specification[5]. Each AC is a variant of DCF with its own parameters AIFSD[AC], CW[AC], CWmin � AC � andCWmax � AC � instead of DIFS, CW, CWmin and CWmax. The AIFSD[AC], which is usually referredto as Arbitration Inter-Frame Space (AIFS), is calculated as AIFSD � AC � � SIFS � AIFS � AC � �SlotTime, where AIFS[AC] is an integer greater than zero. Furthermore, the BC is chosen from [1,1+CW[AC] ] rather than [0, CW] as in the DCF. An AC with higher priority is assigned smallerCWmin and CWmax values and/or shorter AIFS. The IEEE 802.11e EDCF access method is shownin Figure 4.

AIFSD[AC]+ SlotTime

AIFSD[AC]

PIFS

SIFSBusy

Medium

Defer AccessSlot Time

Select Slot and decrement backoffas long as medium stays idle

Immediate access whenmedium is idle >=

AIFSD[AC] + SlotTime

Backoff Window Next Frame

Contention Windowfrom [1, CWmin[AC] +1]

Figure 4: 802.11e EDCF access method

Each AC within a station behaves as a virtual station and contends for Transmission Oppor-tunities (TXOPs), a time interval when it can initiate transmissions, using its own parameters andits own BC. If the BCs of two or more ACs within a station reach zero at the same time, the framefrom the AC with highest priority receives the TXOP. ACs with lower priority behave as if therewere an external collision in the medium and perform backoff process with increased CW val-ues. The four ACs of a 802.11e station and one priority of a legacy 802.11 station and is shown inFigure 5. The AP determines and broadcasts the parameters for each AC plus the limit of a TXOPinterval for each AC TXOPLimit[AC] in beacon frames periodically. During a TXOP, multipleMAC Protocol Data Units (MPDUs) from the same AC in a station can be transmitted with anSIFS time interval between an ACK and the next frame transmission. This multiple transmissionof MPDUs is called Contention Free Burst (CFB).

8


Backoff

AIF

S[0]

BC

[0]

Backoff

AIF

S[1]

BC

[1]

Backoff

AIF

S[2]

BC

[2]

Backoff

AIF

S[3]

BC

[3]

Virtual Collision Handler

Transmission Attempt

Access Category 0 1 2 3

Backoff

DIF

SB

C

Legacy station: single priority

Transmission Attempt

802.11e station: four Access Categories/multiple priorities

Figure 5: 802.11e EDCF queues vs 802.11 DCF queue per station

3.2 Hybrid Coordination Function

The HC, which is usually located at the AP, may assign TXOPs to itself or other stations afterthe medium is idle for PIFS without any backoff procedure, therefore it has higher priority thanACs since ACs, using (E)DCF cannot contend for medium access at least after DIFS idle duration.The HC traffic delivery and TXOP assignment may be scheduled during both CP and CFP. Thebeacon frame is broadcast by the HC at the beginning of each superframe. During CFP, only HCcan allow access to the medium by giving TXOPs to stations using QoS CF-Poll frames, whichspecify the starting time and maximum duration of each TXOP. The CFP ends if the durationreaches the time specified in the beacon frame or the HC sends a CF-End frame. During CP,each TXOP begins when a station gains access to the medium using EDCF or when this stationreceives a QoS CF-Poll frame from the HC. A Controlled Access Period (CAP) is composed ofseveral intervals within one CP when short bursts of frames are transmitted using polling. Atypical 802.11e superframe is shown in Figure 6.

4 Distributed Approaches Based on DCF

In the distributed approach, each station in a WLAN determines to access the medium withoutcontrol of a particular station. Service differentiation between traffic classes is based on differen-tiation of the time the traffic has to wait before transmission. Two main parameters that decidethe waiting time of traffic are Inter-Frame Space (IFS) and CW. The distributed approach may bedivided further into priority-based approach and fair scheduling approach. The priority-basedapproach and fair scheduling approach are described in Section 4.1 and Section 4.2, respectively.

9


Beacon

QoS CF- Poll

Polled TXOP (CFP)

QoS CF- Poll

CFP (polling through HCF)

Transmittedby HC

Transmittedby (Q) STAs

TXOP

TXOP

Polled TXOP (CAP)

RTS/CTS/fragmented DATA/ ACK (polled by HC)

RTS/CTS/fragmented DATA/ ACK (polled by HC)

RTS/CTS/DATA/ACK (after DIFS +backoff)

CF- End

CP (Both EDCF and polling through HCF)

802.11e periodic superframe

TBTT TBTT

Figure 6: A typical 802.11e superframe

4.1 Approaches based on priority

In the priority-based approach, the priority is mapped into medium access. Traffic with higherpriority is assigned smaller values for IFS and/or CW (leading to smaller BI value), thereforelow-priority has to wait longer than high-priority before transmission.

4.1.1 Approaches based on IFS

Aad et al. [7] recommend a scheme in which higher priority j � 1 and lower priority j have IFSvalues of DIFS j � 1 and DIFS j, respectively, such that DIFS j � 1

� DIFS j. The maximum ran-dom range RR j � 1 of priority j � 1 is defined as the maximum BI of that priority. If the strictcondition RR j � 1

� DIFS j � DIFS j � 1 is satisfied, then all packets of priority j � 1 have beentransmitted before any packet of priority j is transmitted. In less stringent condition, RR j � 1 �DIFS j � DIFS j � 1, a packet which could not access the medium the first time may have its pri-ority decreased in the subsequent attempts. Simulations were carried out and the results showthat the method does not change the system efficiency, with data rate sums remain the same. Themethod works well for both Transmission Control Protocol (TCP) and User Datagram Protocol(UDP) flows with more significant effect on UDP flows than on TCP flows. It also works in noisyenvironment and keeps the same stability of the system.

Benveniste [15] suggests a technique to differentiate services based on Urgency ArbitrationTime (UAT), which is the time a station has to wait before a transmission attempt following aperiod when the medium is busy. AIFS and Backoff Counter Update Time (BCUT) are general-ization of DIFS and SlotTime, respectively. Traffic with higher priority is assigned shorter AIFSand/or shorter BCUT values. The highest priority has AIFS

�high prio � � PIFS and a mini-

mum backoff time of 1 in order to prevent conflict with medium access by centralized protocolPCF. Simulation of two traffic classes with AIFS

�high prio � � PIFS, AIFS

�low prio � � DIFS,

CW�high prio � � � 1, 32 � and CW

�low prio � � � 0, 31 � was conducted. The outcomes reveal that

the delay and jitter of high-priority traffic are decreased and under moderate load condition, theperformance of low-priority traffic is also improved as compared with legacy DCF.

Deng et al. [19] propose a method to support two priorities. High and low priorities have towait for the medium to be idle for PIFS and DIFS, respectively, before initiating backoff proce-dure. Simulation of video, voice and data traffic with priorities of 3, 2 and 0, and traffic ratios of 1: 1 : 2, respectively, was performed. The results demonstrate that the approach, when combinedwith separation of random backoff time, can be used to support video, voice and data traffic inheavy load condition (say 90%). In heavy load condition, the highest priority video traffic usesmost of the bandwidth (55%) and lower priorities use the remaining bandwidth, while in light

10


load condition, the lower priority traffic has the required bandwidth. It is also illustrated thatvideo and voice traffic have lower access delay and lower packet loss probability than in DCF,but data traffic has higher access delay and higher packet loss probability than in DCF.

4.1.2 Approaches based on CW

Separation of CW

Aad et al. [8] introduce a differentiation mechanism based on CWmin separation, in which higherpriority traffic has lower CWmin value. Simulations of a WLAN consisting of an AP and threestations with CWmin values of 31, 35, 50 and 65, respectively, were conducted with both TCP andUDP flows. The results reveal that for the same set of CWmin values, the differentiation effect ismore significant on UDP flows than on TCP flows. The per-flow differentiation is introduced, inwhich the AP sends back ACK packets with priorities proportional to priorities of the destina-tions. In other words, the AP waits for a period of time which is proportional to the delay from adestination before transmitting an ACK packet to that destination.

Barry et al. [12, 49] recommend using different values of CWmin and CWmax for different pri-orities, in which higher priority has lower CWmin and CWmax values than those of lower priority.Simulations of high-priority traffic with CWmin between [8,32] and CWmax

� 64, and low-prioritytraffic with CWmin between [32,128] and CWmax

� 1024 were performed. The outcomes showthat the high-priority and low-priority traffic undergo different delay.

Deng et al. [19] propose a scheme based on separation of CW. Originally, the random BIis uniformly distributed between � 0, 22 � i � 1 � , in which i is the number of times the station at-tempted transmission of the packet. In Deng’s scheme, the high and low priorities have randomBI values uniformly distributed in intervals � 0, 22 � i � 2 � 1 � and � 22 � i � 2, 22 � i � 1 � , respectively.This approach can be combined with the approach based on IFS mentioned in the last paragraphof Section 4.1.1. The simulation results reveal some improvement in delay and jitter for highpriorities (voice and video).

Xiaohui et al. [53] present the Modified DCF (M-DCF) scheme, which uses different valuesof CWmin and CWmax for service differentiation. Simulations of ad-hoc WLAN with 10 data sta-tions and between 10 and 35 voice stations were performed. Voice service had CWmin

� 7 andCWmax

� 127, while data service had CWmin� 15 and CWmax

� 255. The outcomes illustratethat M-DCF decreases the total packet dropping probability and the dropping probability of voicepackets as well as reduces the contention delay of both voice and data packets as compared withDCF.

Different Priority/Persistent/Persistence Factor (PF)

Aad et al. [6, 7] propose a method based on the backoff increase function. In the original DCF,the CW is multiplied by a Priority Factor (PF) of 2 after each collision. In Aad’s method, a higherpriority traffic has a lower PF Pj. Simulations of three priorities with PF values of 2, 6 and 8 wereconducted. The results demonstrate that this method works well with UDP flows, but does notwork well with TCP flows or in noisy environment. The efficiency is not lost but the stability ofthe system is decreased.

Benveniste [15] recommends a technique based on Persistent Factor (PF). After each collision,the CW is multiplied by a PF. Higher priority traffic has lower value for PF. For time sensitiveapplications and with capability for congestion estimation, PF value should be less than 1. Oth-erwise, a value between 1 and 2 should be chosen. Simulation of traffic with two priorities andAIFS

�high prio � � PIFS, AIFS

�low prio � � DIFS, PF

�high prio � � 0.5 and PF

�low prio � � 2

was carried out. The outcomes show that the high-priority traffic performance is improved with-out any significant effect on low-priority traffic. Furthermore, the delay and jitter is less than 10ms, which could not be obtained with differentiation based on IFS alone.

Song et al. [46] introduce a scheme called Exponential Increase Exponential Decrease (EIED),in which after a collision or a successful transmission, the CW is multiplied or divided by factor

11


rI and rD, respectively. Simulations of 5, 10, 40 and 60 stations with packet arrival rate between10 to 160 packets/s and packet length of 1024 bytes were performed using the Binary ExponentialBackoff (BEB) and EIED with [rI , rD] = [2, 21 � 8], [2, 21 � 4], [2 � 2, 2 � 2], [2, 2]. The results revealthat the delay of EIED is smaller than that of BEB for all packet arrival rates and all number ofstations. The throughput of EIED is the same as throughput of BEB when the packet arrival rateis small (less than 80 packets/s) and higher than throughput of BEB when the packet arrival rateis high.

4.1.3 Approaches based on combination of IFS and CW

Wong et al. [52] present a scheme named Age Dependent Backoff (ADB), which is an improve-ment of EDCF. For each TC, after each collision, the CW[TC] is multiplied by a Persistence Factor(PF) given by the formula

PF � TC � � � 2LT � TC � � Age � 2

LT � TC � and Age are the packet lifetime (LT) and age of the packet in transmission queue, respec-tively. Packets with Age � LT are discarded since they would be obsolete by the time they get tothe recipient. PF � TC � is between 1 and 2 in the first half of packet lifetime and is between 0 and1 in the second half. CW � TC � is always less than CWmax � TC � but could be less than CWmin � TC � todifferentiate between a retransmitted packet and a newly arrived one. Simulations of an ad-hocnetwork with 10 voice stations, 4 video stations and a number of File Transfer Protocol (FTP)stations (between 5 and 25), in which half of the FTP stations using DCF and the other half usingEDCF, were conducted. LTs of voice and video packets were set to 25 ms and 75 ms, respectively.DCF had DIFS � 50µs, CWmin

� 31, CWmax� 1023. For voice service, AIFS � 1 � � Voice � � 50µs,

CWmin � 1 � � 7, CWmax � 1 � � 31. The video service had AIFS � 2 � � Video � � 70µs, CWmin � 2 � � 15,CWmax � 2 � � 63, while data service had AIFS � 3 � � Data � � 110µs, CWmin � 3 � � 15, CWmax � 3 � � 255.The results show that ADB improves the packet delay, jitter and drop rate of both voice and videotraffic as compared with EDCF with fixed PF values of 2 and 1.5. At the same time, the best-effortFTP traffic does not suffer from starvation.

Romdhani et al. [40] introduce a method called Adaptive EDCF (AEDCF), in which CW isdynamically calculated after each successful transmission or collision according to the networkconditions. After each successful transmission of a packet of class i, the CW is updated as

CWnew � i � � max�CWmin � i � , CWold � i � � MF � i � �

where higher priority has lower multiplicator factor MF � i � . If MF � i � is fixed at 0.5, then thescheme is called Slow Decrease (SD) scheme. Each traffic class has a different multiplicator factor(MF), which should not exceed 0.8 as suggested by the authors.

MF � i � � min� �

1 � �i � 2 � � � f j

avg, 0.8 �The average collision rate at step j is

f javg

� �1 � α � � f j

curr�

α � f j � 1avg

where α is the weight or smoothing factor. The average collision rate is calculated every Tupdateexpressed in SlotTime. The estimated current collision rate is

f jcurr

� E�collision j � p � �

E�data sent j � p � �

where E�collision j � p � � and E

�data sent j � p � � are the number of collision and the number of packets

sent at station p in the period j, respectively. After each collision, CW of class i is updated as

CWnew � i � � min�CWmax � i � , CWold � i � � PF � i � �

12


where higher priority traffic has lower persistent factor PF � i � . Simulations of an ad-hoc WLANwith the number of stations between 2 and 44 (corresponding to load rate from 6.7% to 149%)were carried out. Each station had three service classes, with high, medium and low classes hav-ing PF values of 2, 4 and 5, respectively. Tupdate

� 5000 � SlotTime and α � 0.8. High, mediumand low classes had AIFS values of 34µs, 43µs and 52µs, respectively. � CWmin, CWmax � of high,medium and low classes were [5,200], [15,500] and [31,1023], respectively. The outcomes demon-strate that AEDCF has lower mean access delay than EDCF and SD schemes, and its delay is lessthan 10 ms. AEDCF and SD has higher goodput than EDCF. Furthermore, AEDCF has highergain on goodput than SD, especially under high load condition. The medium utilization de-creases when traffic load increases, but AEDCF has higher utilization than SD, which has higherutilization than EDCF. Of all the three schemes, AEDCF has lowest collision rate while EDCF hashighest collision rate.

Kim et al. [29, 30, 31] propose a technique called DCF with Shorten Contention Window(DCF/SC), which provides support for three service classes, premium (low latency and jitter),assured (guaranteed bandwidth) and best-effort (no guarantee). The super period, which orig-inally consists of PCF and DCF, is divided into two periods. In the first period, only premiumservice is allowed to access the medium with DCF/SC after sensing the medium idle for PIFS.In the second period, premium and assured services access the medium with DCF/SC and best-effort service access medium with DCF after medium idle time of DIFS. DCF/SC has shorterCW than original IEEE 802.11 DCF. Simulations of infrastructure WLAN with premium servicesuch as voice, Motion Picture Extreme Compressed Movies (MPX) and Motion Picture ExpertsGroup 2 (MPEG2), assured service (assured bulk data) and best-effort service (bulk data) wereconducted. DCF/SC had CW values from [11, 17, 23, 29, 35, 41] while DCF had CW values from[16, 32, 64, 128, 256, 512, 1024]. The results show that the method increases utilization (definedas the ratio of total successful transmission time of pure data (excluding backoff time, preamblesand header) to the total simulation time) and increases stream throughput (defined as the ratioof the number of successful packets to the generated packets) as compared with the DCF. It alsodecreases the latency and jitter of the real-time services.

Banchs et al. [11] suggest a scheme for differentiation of real-time and best-effort services.Real-time stations waits for medium to be idle for PIFS and generates elimination burst EB 1,whose duration is multiple of SlotTime. The length of EB 1 has the following distribution

PE1�n � �

�pn � 1

E1 � � 1 � pE1 � , 1 � n � mE1

pmE1 � 1E1 , n � mE1

where n is the length of EB 1 in SlotTime, pE1 is a probability between 0 and 1 and mE1 is themaximum length of EB 1 in SlotTime. If the station senses the medium busy after transmissionof EB 1 or EB 2, it defers access to the medium until the medium is idle for PIFS. The station withthe longest EB 1 or EB 2 gains access to the medium. If two stations have the same EB 1 and EB 2,collision occurs and would be detected. Data stations wait for the medium to be idle for DIFS anduse backoff algorithm. Each station has a share value corresponding to QoS level of this station.The basic service has share value of 1 and higher service has share value greater than 1. For allstations, the ratio wi

� risi

should be equal, where ri and si are throughput and the share assignedto station i, respectively. ri is updated after a packet transmission as follows

rnewi

� �1 � e �� ti � k � � li�

ti

� e �� ti � k � roldi

where li and�

ti are the length and inter-arrival time of the transmitted packet, and k is a con-stant. Each station calculates its own wi and includes this in the transmitted packet’s header. Astation observes a packet and notices the value of wi in this packet header. If the station’s wi isgreater than this packet’s wi, the station increases its CW by a small amount, and vice versa. Thealgorithm for calculation of a station CW is

if�wown � wrcv � then

13


CW ��1 � �

1 � � CWelse if (queue empty) then

CW ��1 � �

1 � � CWelse

CW ��1 � �

1 � � CWend if

where CWMin802.11 � CW � CWMax802.11 and wown is calculated by the station, wrcv is the valueof wi in the header of observed packet,

�1 is calculated as

�1

� k��

wown � wrcv

wown� wrcv

��

where k is another constant, which is different from k in the formula for rnewi . It is claimed that

the residual collision rate (two or more stations collide after transmission of EB 1 and EB 2) isvery small and only depends on pE1, mE1 and mE2, but not the number of contending stations.

Sheu et al. [41, 42] present a scheme called Distributed Bandwidth Reservation Protocol(DBRP) to support both real-time and non real-time traffic. Data stations use DCF: they waitfor the medium to be idle for DIFS and start backoff algorithm with CWmin and CWmax values of31 and 1023, respectively. Voice stations stay in one of three states: initial state where each sta-tion starts with, reservation state where voice stations contend to reserve access to the mediumand transmission state where voice stations transmit periodically without contention. A voicestation has an Sequence ID (SID) for access sequence and an Active Counter (AC) for the num-ber of active voice stations. A voice station that wants to access the medium at time t sensesthe medium for the reservation frame (RF) in period

�t, t � Dmax � , where Dmax is the maximum

acceptable delay of voice packets. The RF stores the number of active voice stations (AN). Ifthere is no RF frame, and if the medium is idle in the period

�t � Dmax, t � Dmax

� PIFS � , thevoice station performs backoff algorithm with backoff time uniformly distributed between 0 and3 � SlotTime. When the backoff time goes to zero, the station enters Send RTS procedure and trans-mits RTS frame to destination. If there is no collision, the station would receive CTS frame, entersTransmission State and becomes the Reservation Cycle Generator (RCG), which is the first voicestation in WLAN and has to generate RF frame periodically. It sets its SID and Active Counter to1 and transmits RF frame and its voice packet. If there is RTS collision, the colliding voice stationwould use p-persistent scheme for retransmission of RTS, with probability pp in the next timeslot. If there is RF frame in period

�t, t � Dmax � , the voice stations enters RF received procedure,

sets its Active Counter as the value of AN in the RF frame (denoted as RF.AN). The voice stationenters Wait to content procedure and waits for a duration of WT � RF.AN � Tvoice before trying toaccess the medium. If the medium is idle for SlotTime during this period, a voice station has leftand WT should be decreased by one Tvoice. After WT, the voice station goes to backoff procedureto send a RTS frame. The condition Dmax

�RPmax

� TmaxMPDU is required, where RPmax is thesum of the maximum voice packets reservation period and the voice station contention period,TmaxMPDU is the transmission time of a maximum MPDU. If there is no collision of RTS/CTS,the station increases its Active Counter by 1 and sets its SID to the content of Active Counter.The station sends the first packet and enters Transmission State. All other stations increase theirActive Counter by 1 when they hear a CTS. If there is collision of RTS, the colliding stations usep-persistent scheme for contention resolution. After the last access that is over the boundary ofRPmax, a voice station senses the medium idle for PIFS. If the voice station SID is 1, it is the RCGand sends the RF frame and the voice packet immediately. Other stations sense the medium, andif it is idle for SlotTime, the RCG has left and all stations decrease its SID and Active Counterby 1. The new RCG has to transmit the RF frame before its voice packet. When an RF frameis received, a voice station sets its cd timer as cd timer � �

SID � 1 � � Tvoice. When the cd timerreaches zero, the station transmits its packet.

Chen et al. [16] recommend a method to differentiate between real-time and non-real-timetraffic based on IFS, CWmin and retransmission procedure of real-time packets. After the mediumis idle for DIFS, stations with real-time packets for retransmission generate jamming noise. The

14


station with longest jamming noise then transmits its packet. The length of jamming noise has atruncated geometric distribution. The probability of a jamming noise with length of f in unit ofSlotTime is:

Pj�f � �

�p f � 1

j � � 1 � p j � , 1 � f � JW

pJW � 1j , f � JW

p j is a probability value between 0 and 1, Jamming Window JW is the maximum jamming timedefined as JW � min

�JW � p JW � 1

j � 1N � , where N is the number of stations that simultaneously

generate jamming noise. Simulations of seventy non-real-time flows and a number of active real-time flows were performed using Chen’s scheme, Tiered Contention Multiple Access (TCMA)[13, 14] and Virtual DCF (VDCF) [17, 18]. (CWmin, IFS) for real-time and non-real-time were(15, 40 µs) and (31, 50 µs), respectively. p j was 0.35 and JW was 9. The results show that Chen’sscheme has lower mean MAC delay, lower packet dropping rate, similar average jitter and higherchannel utilization than TCMA. This scheme also has lower mean MAC delay, similar packetdropping rate, lower average jitter and lower channel utilization than VDCF.

Sobrinho et al. [43, 44, 45] propose a scheme called Blackburst. Define τ to be the maximumpropagation delay between any two stations. There are three inter-frame spacings that are re-quired for the access procedures of data and real-time stations, which must satisfy the conditionstshort

� 2 � τ� tmed and tmed

� 2 � τ� tlong. A data station that wants to transmit data senses

the medium for a period tlong. If the medium is idle, this station transmits data. Otherwise, itwaits for the medium to be idle for tlong and enters the backoff procedure, with the BI fdata

�c � in

unit of white slot twslot asfdata

�c � � rand � fdata

�0 � � 2c �

where rand � a � returns a random number between 0 and a � 1, c is the number of collisions thatthe data packet has experienced and fdata

�0 � is the initial CW. A real-time station that wants to

transmit packets senses the medium for a period tmed. If the medium is idle, this station transmitsa packet. Otherwise, it waits until the medium is idle for tmed again and enters the blackburst con-tention period. The station jams the medium with a black burst in unit of black slot tbslot, whoseduration is proportional to the time that this station has been waiting. After the transmission ofthe black burst, the station waits for tobs to see if any station is transmitting a longer black burst.The winning station, which is the station that has been waiting the longest time, would get ac-cess to the medium and other stations would contend for the medium with blackburst in the nextcycle. In every black burst contention period, there is only one unique winner. By ensuring that2 � τ

� tobs� min

�tbslot, tmed � , a real-time station can always see if its blackburst is shorter than

that of another and real-time stations do not attempt to access the medium during the observa-tion interval. When a real-time station has access to the medium, it transmits a packet that mustbe at least tpkt and schedules the next access instant to be tsch in the future. The real-time stationshave higher priority than the data stations, since no data station would perceive the medium idlefor tlong until all real-time stations have access to the medium. The real-time stations get access tothe medium in the round robin order and share the medium in a Time Division Multiple Access(TDMA) manner. After all the real-time stations have synchronized their transmissions, the blackburst contention is only initiated again if some transmission of data stations disturbs the order.

4.1.4 Discussion

Aad’s scheme [7] can support a large number of priorities with the proper selection of DIFS jfor each priority. Benveniste’s approach [15] of using BCUT for differentiation is not possiblebecause the higher priority could not use a shorter BCUT as the SlotTime used in the originalDCF is the minimum possible. Deng’s method [19] uses only two IFS values, which are PIFS andDIFS, therefore it only allows for differentiation of two priorities. In order to support more thantwo priorities, more IFS values must be used. However, except PIFS, all other values of IFS areat least DIFS and therefore there can be only one higher priority than the legacy DCF traffic ifdifferentiation is merely based on IFS values. Furthermore, the total time a station has to wait

15


before gaining access to the medium is the sum of IFS and the random backoff time. Therefore,even if high-priority traffic has lower IFS value, it can still have higher total wait time than low-priority traffic and loses the medium contention to low-priority traffic.

Although Barry’s, Deng’s and Xiaohui’s methods [12, 19, 53] only provide service differen-tiation for two priorities, the same principle can be applied to support more priorities. Song’sscheme [46] shows improvement in delay and throughput compared with DCF, however it doesnot introduce differentiation of services. Aad’s methods [6, 7] based on IFS or backoff increasefunction does not work well for TCP flows, since the AP always uses its own priority to sendback TCP-ACKs to different stations. For example, if two stations waits for t1 and t2 before trans-mitting a packet on average, then the data rate ratio is t2

� t1. For TCP flow, the ACK producesadditional delay t0 and the data rate becomes

�t2

� t0 � � � t1� t0 � . If the AP is slow, t0 is high and

ratio�t2

� t0 � � � t1� t0 � is very much different from the desired ratio t2

� t1 [8]. In using differentCWmin and/or CWmax values for different priorities, low-priority traffic has to produce longerbackoff time even if there is no high-priority traffic, leading to longer delay. Choosing a randomBI in the CW range produces unpredictable variations in delay and throughput, which is unde-sirable for time-sensitive traffic. With regard to the PF, if PF value is greater than 1, after eachcollision, the CW is increased and the collided packet has to wait longer before being retrans-mitted, resulting in longer delay and higher jitter, which is not desirable for real-time packets.However, if PF value is less than 1, CW is reduced, and under the heavy load condition, conges-tion may increase leading to more collisions.

In Sheu’s method [41, 42], DIFS � PIFS � 4 � SlotTime, which is inconsistent with the cur-rent standard, in which DIFS � PIFS � SlotTime. Sobrinho’s method [43] can provide guaranteeon delay. However, it is only optimized for isochronous traffic and therefore can be a drawbackfor variable rate traffic [49]. Bandwidth is also wasted by sending bursts for each packet to accessthe medium [42]. The approach using combination of IFS and CW can support more prioritiesthan each of the IFS or CW approach alone. However, the values of IFS, CW and BI are determin-istic for each priority once chosen. This may lead to unfair medium access since high prioritiestend to seize the medium and low priorities may suffer starvation. Romdhani’s and Wong’smethods [40, 52] calculate the CW dynamically based on network conditions, and therefore canadapt better to traffic load and can avoid starvation of low-priority traffic.

4.2 Approaches based on fair scheduling

Fair scheduling approach guarantees that traffic in each class has a fair chance for transmission.The approach aims to ensure that the bandwidth given to traffic flows is proportional to theirweights.

4.2.1 Matching priority to IFS

Pattara-Aukom et al. [37] propose a method called Distributed Deficit Round Robin (DDRR)based on Deficit Round Robin (DRR) scheduling. Traffic is categorized into classes. Traffic classi has a service quantum Q bits every Ti seconds, such that Q � Ti is equal the desired throughputof class i. The Deficit Counter (DC) of traffic class i, at MS j at any given time is DC j

i and is

proportional to the bandwidth available to this traffic class at that time. DC ji is increased Q bits

every Ti seconds and is decreased by the size of the frame whenever a frame is transmitted. Attime t, IFS for traffic class i at MS j, IFS j

i�t � is calculated as

IFS ji�t � � DIFS � α � DC j

i�t �

Q� random

�1.0, β �

DC ji�t � � DC j

i�t � � � Q

Ti� � t � t � �

DC ji�t � � DC j

i�t � � Frame Size

�t �

16


α is a scaling factor, β � 1 and random�1.0, β � returns a random number uniformly distributed

between 1 and β.

0 � DC ji � DIFS � PIFS

α� Q

If DC ji

� 0, then traffic class i has to wait until DC ji is greater than 0 again before next transmis-

sion. The above equations leads to

PIFS � IFS ji � DIFS

From this last equation, DDRR is backward compatible with IEEE 802.11 When a MS senses themedium is idle, it waits for IFSi before transmitting a traffic class i frame. If the medium is busy,the MS waits until it becomes idle again, waits for additional time IFSi (as calculated at this time)and then transmits the frame. There is no backoff procedure in this scheme. The right to accessthe medium depends on DC j

i but does not depend on the required throughput, resulting in fairshare of the medium between different traffic classes.

4.2.2 Matching priority to CW

Banchs et al. [9] present an approach called Distributed Weighted Fair Queuing (DWFQ), inwhich for all stations, the ratio Li

� riWi

should be equal, where ri and Wi are throughput andthe weight assigned to station i, respectively. The assumption is all packets at a station are fromone flow. IEEE 802.11 stations have default weight of 1, and other weight values must be greaterthan or equal to 1, which means better than or best-effort service. ri is updated after a packettransmission as follows

rnewi

� �1 � e � ti � K � � li

ti

� e � ti � K � roldi

where li and ti are the length and inter-arrival time of the transmitted packet, and K is a constant.Each station calculates its own Li and includes this in the transmitted packet’s header. A stationobserves a packet and notices the value of Li in this packet header. If the station’s Li is greater thanthis packet’s Li, the station increases its CW by a small amount, and vice versa. The algorithmfor calculation of a station CW is

if (overload) thenp �

�1 � �

2 � � pelse if

�Lown � Lrcv � then

p ��1 � �

1 � � pelse if (queue empty) then

p ��1 � �

1 � � pelse

p ��1 � �

1 � � pend ifp � min

�p, 1 �

CW � p � CW802.11

where�

2 is a constant, Lown is calculated by the station, Lrcv is the value of Li in the header ofobserved packet, p is a scaling factor and

�1 is calculated as

�1

� k��

Lown � Lrcv

Lown� Lrcv

��

where k is another constant, which is different from k in the formula for rnewi . The condition

overload occurs when there are a large number of stations with high weights, resulting in stationschoosing small values for CW and there are a large number of collisions. This condition could betested like this

if (av nr coll � c) then

17


overload � trueend if

where c is a constant that needs to be chosen properly. It is a trade-off between the efficiency of themedium and the closeness to the desired bandwidth allocation. If c is too small, flows with highweights may not reduce their CWs adequately resulting in insufficient bandwidth allocation. If cis too high, the number of collisions is high and the medium efficiency is decreased. The averagenumber of collisions is updated after each successful transmission as follows

av nr coll � �1 � t � � num coll � t � av nr coll

where av nr coll in the right hand side of the equation is the last calculated value of averagenumber of collisions and t is a small smoothing factor. For the case where a node i transmits nflows with weights W1, . . . , Wn, it uses the label

Li� ri

∑nj � 1 Wj

where ri i the total bandwidth of the node. Then this node uses weights W1, . . . , Wn to choose thenext packet in its flows for transmission.

Banchs et al. [10] suggest a scheme called Assured Rate MAC Extension (ARME), in whichafter each packet transmission, if the estimated sending rate of the station is higher (or lower)than the desired rate, the station CW is slightly increased (or decreased). The algorithm for CWcalculation is as follows

if�overload � then

p ��1 � �

4 � � pelse if

�qlen � 0 � then

p ��1 � �

1 � � pelse if

�bsize � blim � then

p ��1 � �

2 � � pelse

p ��1 � �

3 � � pend ifp � min

�p, 1 �

CW � p � CW802.11

where�

4 is a constant, qlen is the queue length and CW is only decreased when this is greaterthan 0; blen is the bucket length of the resource that the station has for transmission; blim is theminimum length required for the bucket resource before a transmission occurs;

�1 is a constant

and�

2 and�

3 are calculated as follows

�2

� blim � blenblim

� �1

�3

� blen � blimbsize � blim

� � 1

where bsize is the acceptable burstiness of the source, with the maximum burst length equalbsize � blim. The condition overload occurs when there are a large number of stations with highweights, resulting in stations choosing small values for CW and there are a large number ofcollisions. This condition can be tested as mentioned in the previous paragraph.

Qiao et al. [39] recommend a priority-based MAC (P-MAC) protocol. There are n traffic classeswith weights such that 0 �

φn� . . . �

φ2�

φ1� 1. Assume that each station only has one

traffic flow. Let fi be the set of stations that carry class traffic i. Assume that the traffic flows havethe same MAC frame size. Then the weighted fairness is achieved if

�i, j �� 1, . . . , n � , � s � fi,

�s � � f j,

SUs

φi

� SUs �φ j

18


where SUs is the probability that station s transmits a frame successfully. This is equivalent to�

u, v � fi, CWu� CWv

and �j �� 2, . . . , n � , CW j

� CW1 � 1φ j

� 1

In P-MAC, the CW of each station is chosen to achieve the weighted fairness shown above andto reflect the number of stations in contention for medium access in order to maximize aggregatethroughput. The number of traffic classes and their weights are assumed to be known in advanceand CW1 is set to a given initial value. Let avg idle and avg wait i be the average number of con-secutive idle slots and the average number of slots between two adjacent successful transmissionof traffic class i, respectively. The stations sense the medium to determine at each slot whetherthe medium is idle or busy, if the busy period is due to collision or successful transmission of apacket of which traffic class. Let an idle-busy-cycle be the duration between the end of a busyperiod to the end of the next busy period. After each idle-busy-cycle, avg idle is updated whileavg wait i is only updated if a traffic class i packet is successfully transmitted. The calculationof these values uses the corresponding previous average value with a smoothing factor α. Atthe end of each observation window, the � fi � is updated and avg idle and avg wait i are reset.The calculation of � fi � uses the previous value with smoothing factor β, where the instantaneousvalue is given by

� fi � ��CWi � 1 � � � avg idle � 1 �

2 � avg idle � � avg wait i � 1 �The optimal value for CW1 is calculated base on these values of � fi � and the optimal CW valuesfor other traffic classes are updated according to the formula above.

Wang et al. [21, 50] suggest a method in which each station considers all other stations asa whole entity and therefore only has the notation of itself and the others. Each station has apre-defined fair share of the bandwidth that it should receive. When a station receives a packet,depending on the type of packet, it updates estimation of its share of bandwidth or the others’share of bandwidth accordingly. If a station receives a packet destined for it or for the others, itupdates estimation of its own share of bandwidth or the others’ share of bandwidth, respectively.If the optional RTS/CTS mechanism is used (this is obvious in reception of a RTS/CTS packet,however, in case a DATA or ACK packet is received, the corresponding DATA packet size wouldbe greater than the RTS THRESHOLD), the RTS/CTS packets are also included in the estimationof share of bandwidth. If a station receives a RTS, CTS or DATA packet destined for the others,it updates the transmission time of the corresponding DATA packet. When this station receivesan ACK packet destined for the others, the latest transmission time value of the DATA packetdestined to the others is used in calculation of the others’ share of bandwidth. Define the esti-mated fairness index FIe

� Wei�i

� Weo�e

, where Wei and Weo are estimated throughput of station iand the others, respectively.

�

i and�

o are the pre-defined fair share of station i and the others,respectively. If the station estimates that it has more than/enough/less than the required shareof bandwidth (FIe is greater than C/between 1

C and C/less than C, where C is a constant greaterthan but close to 1) it would double/keep/halve the current CW until the CW reaches the CWmaxor CWmin.

4.2.3 Matching priority to BI

Vaidya et al. [48] introduce the Distributed Fair Scheduling (DFS) algorithm based on Self ClockFair Queuing (SCFQ) and IEEE 802.11 DCF. The BI of a packet is chosen proportional to its finishtag and therefore the packet with smallest finish tag would be transmitted first. Assume eachstation only has one flow. Each packet has a start tag and a finish tag. When a station i hearsor transmitted a packet at time t with finish tag Z, it sets its virtual clock to maximum

�vi�t � , Z � ,

19


where v(t) is the virtual time at real time t. When packet Pki reaches the front of the queue at

station i at real time f ki , it is marked with a start tag Sk

i� v

�f ki � . Its finish tag is calculated as

Fki

� Ski

� Scaling Factor � Lki

φi

where Scaling Factor is for suitable scale of virtual time, Lki is the length of packet and φi is the

weight of station i. At time f ki , this packet is also assigned a backoff interval in unit of SlotTime

Bi�� Fk

i � v�f ki ��

where � a � returns the maximum integer not greater than a. For collision reduction, Bi is ran-domized with a random variable ρ uniformly distributed in [0.9 1.1]. The final expression for Biis

Bi�� ρ � Scaling Factor � Lk

iφi�

If there is a collision, the colliding stations choose a new backoff interval uniformly distributedbetween � 1, 2CollisionCounter � 1 � CollisionWindow � , where CollisionCounter stores the number ofcollisions that this packet has experienced and CollisionWindow is a constant. If a station hasmultiple flows, when a packet reaches the front of a flow, its start tag and finish tag are marked.A station chooses the packet with the smallest finish tag among the packets at the front of itsflows. The BI for this packet is then calculated as described above.

Dugar et al. [20] propose an approach combining the ideas from DFS and High PerformanceLAN/1 (HIPERLAN/1) to match packet length, priority and weight of a flow to which the packetbelongs, to BI. Station i with weight φi chooses BI Bi for its kth packet Pk

i with packet length Lki as

follows

Bi�� ρ � Scaling Factor � Lk

iφi�

where ρ is a random variable uniformly distributed between [0.9, 1.1], Scaling Factor is usedfor a suitable scale for backoff interval and � a � returns the greatest integer not greater than a.The backoff interval is represented in a base-N format (for example, if backoff interval is 33 indecimal and base N is 2, then the base-2 representation of backoff interval is 100001). The sta-tion constructs a tuple

�p, c, n, d � n � 1 � , . . . , d0 � with the following elements. p is the priority of the

station. If inter-round spacing irs is M slots, then M priority levels from 0 to M-1 could be sup-ported. The traffic with lower priority level has higher priority. c is collision status of a station.Each station keeps a collision counter ccntri, which is incremented on each collision and is resetafter each successful transmission. If ccntri � 0, then c is zero, otherwise c is 1. n is the numberof digits in the base-N representation of the backoff interval. di is the ith digit in the base-N rep-resentation of the backoff interval, with the 0th digit is least significant. When a station senses themedium idle for inter-round spacing irs, it transmits a burst to start contention resolution proce-dure. The irs is chosen greater than the maximum idle time during the contention resolution, sothat a station would not interrupt an ongoing contention resolution cycle. Each element in thetuple represents a phase in the contention resolution cycle. Only stations that survive the firsti-1 phases go to the ith phase. In the ith phase, a station senses the medium for ti slots, whereti is the ith digit from left in the tuple

�p, c, n, d � n � 1 � , . . . , d0 � . If the station hears a transmission,

it drops out of contention. After the last phase, there are potential winner(s) of the contentionresolution cycle. When station i, the winner of the contention resolution cycle, gets permission totransmit, it piggybacks its backoff interval Bi and its priority level in the data packet. A station jwith the same priority level, hearing this packet transmission, recalculates its backoff interval asB j

� �B j � Bi � if B j

�Bi. Station j then reconstructs the tuple from this new, smaller backoff in-

terval. This procedure ensures that flows with the same priority level share the bandwidth fairly.If two or more stations with highest priority choose the same smallest backoff interval, then theyare the winner of the contention resolution cycle and there is a collision. Each colliding station

20


increments its collision counter by 1 and chooses a new backoff interval uniformly distributedbetween � 1, 2ccntri � 1 � CollisionWindow � where CollisionWindow is a constant. The station thenconstructs a tuple from this new backoff interval and waits for the medium to be idle for irs beforecontending with the new tuple.

Ogawa et al. [35] propose a scheme for control of Class of Service (CoS). When the kth packetpk

i reaches the front of the queue at station i, its start tag Ski , finish tag Fk

i and Packet Virtual Time(PVT) Vk

i are calculated as follows�� Ski

� ViFk

i� Sk

i� Scaling Factor � Lk

i� wi

Vki

� Vi

where Lki is the packet size of packet pk

i , wi (0 � wi � 1) and Vi are weight and Station VirtualTime (SVT) of station i, respectively. The initial value of SVT and PVT are set to zero. Whenstation i hears or transmits a packet with its finish tag z, its associated PVT and SVT are set asfollows

Vki

�� max�Vk

i , z � , Fki � z

Vki , otherwise

Vi� max

�Vi, z �

When a packet reaches the front of the queue, its BI in unit of SlotTime is calculated as

Backo f f Time � � � Fki � Vk

i � � Rand� � � 22 � j �

where � a � returns the greatest integer not greater than a, Rand� � returns a random number uni-

formly distributed between (0,1), j is the number of transmission attempts of the packet. The fairthroughput is achieved with the scaling of SlotTime according to weight. Define Re f erence SlotTimeas SlotTime of weight 1, SlotTime of station i with weight wi is SlotTime � Re f erence SlotTime � wi.

4.2.4 Discussion

The principle of fair queuing is to regulate the time that traffic has to wait according to its pri-ority, such that each traffic class has an equal opportunity for transmission and a bandwidthproportional to its priority. A fair queuing approach involves choosing a fair queuing mecha-nism, mapping of user requirements and using the parameters for service differentiation. Themechanism influences protocol complexity and computational cost [37].Pattara-Atikom’s method [37] is based on DRR, which has O

�1 � complexity and provides sim-

ple mapping of QoS to IFS. There is no backoff procedure, therefore the throughput and delayvariation are improved. The desired throughput and medium access right are mapped into sep-arate parameters, service quantum rate and DC, respectively. This solves the fairness problemand prevents starvation of low-priority traffic. Banchs’s approach [9] DWFQ offers a flow withaverage bandwidth according to its weight, however no guarantee can be granted for individ-ual packets. Therefore this approach may show some short-term unfairness. It involves smallchanges in the calculation of CW and leads to legacy 802.11 stations behaving as best-effort sta-tions. Banchs’s method [10] ARME requires small modifications in the calculation of CW andlegacy 802.11 stations receive best-effort service. It can provide absolute throughput guaranteein normal conditions. Qiao’s P-MAC [39] does not consider traffic delay/jitter requirements,therefore it may not be fair in terms of delay/jitter. Wang’s algorithm [50] can solve the fairnessproblem when packet lengths are variable and work with both basic and RTS/CTS access meth-ods. Vaidya’s mechanism [48] DFS is based on SCFQ, which is of complexity O

�log

�n � � , where

n is the number of flows [37]. DFS can provide relative throughput guarantees. Nevertheless, itdoes not consider delay in the differentiation analysis. Each station has to examine the finish tagof every packet and the header format of 802.11 must be changed to include the finish tag [10].The scheme does not address the maximizing of the channel utilization [39].

21


5 Centralized Approaches Based on PCF

In centralized approach, stations obtain the right to access the medium from a certain station.This station coordinates the medium access by polling stations, the order in which stations arepolled depends on the mechanism used.

5.1 Priority-based approaches

Ganz et al. [22] propose a robust SuperPoll method. In the SuperPoll protocols, a list of stationsallowed to transmit in a certain period is announced at the start of that period. The PC determinesthe polling sequence in the current PCF period for all registered stations. It calculates WPC

��GroupSize � � � TDATA

� TACK � , the maximum time that it would wait before ending CFP, whichis at most the maximum CF duration, with TDATA and TACK are the transmission time of data andACK packets, respectively and GroupSize is the number of stations in the initial SuperPoll. At thebeginning of the current PCF period, the PC broadcasts a Beacon and an initial SuperPoll, whichis a sequential polling list with IDs of the registered stations and sets CNTS (the current number ofNothing To Send (NTS) packets sent) to 0. When the WPC expires or the last station in the pollinglist finishes transmission, the PC broadcasts the CF-End to reset the NAV and end the CFP. Allstations register with the PC during DCF period. Each station needs to listen to the PC andothers to find out whether it is in the polling list, its order in the list, its predecessor (prior stationin the polling list) and the maximum time it has to wait before transmission, which is Wmax

��order � 1 � � �

TDATA� TACK � � CNTS �

�TDATA

� TACK� TNTS � , where TNTS is transmission

time of NTS packet. If a station is not in the polling list, it does nothing. If it is the first station inthe polling list (received from the initial SuperPoll or a data or NTS packet), it starts transmission.If it is not the first station, it calculates its Wmax timer and when this timer expires, it sets CNTS to 0and starts transmission. If a station is in the polling list but has no data to transmit, it incrementsCNTS by 1 and sends a NTS packet. A station transmitting (data or NTS packet) during the PCFperiod includes in its data or NTS packet a remaining SuperPoll, which has the initial polling listwithout all the stations prior to this current station in the polling order and the new CNTS value.At the beginning of CFP, when a beacon is received, the station sets the NAV and stops its timer.At the end of CFP, when a CF-End is received, NAV is reset and timer is stopped.

Suzuki et al. [47] present two polling schemes for PCF. In priority scheme 1, the AP addsall multimedia stations to its polling list at the beginning of CFP and polls the stations in roundrobin manner. However, the stations use a static priority scheduling algorithm for transmissionof MPDUs. Voice, video and data MPDUs have high, medium and low priorities, respectively.According to priority scheme 2, at the beginning of CFP, the AP adds all multimedia stationsto its high-priority and low-priority polling lists. If data stations want to transmit data to APusing PCF, they are added to the low-priority polling list. The AP polls stations in the high-priority polling list in round robin. After all stations in this high-priority list finish transmission ofvoice and video MPDUs and are dropped from the high-priority list or the high-priority pollingperiod expires, AP polls stations in low-priority list in round robin manner. When all stations aredropped from the low-priority list or the CFP expires, AP terminates the CFP.

Yeh et al. [54] introduce four polling schemes for PCF. In the Round Robin (RR) scheme,the PC polls the stations in turn, starting from the station with lowest address until the CFPduration reaches its maximum. The types of message from PC to stations include: CF-Poll, CF-Ack-Poll, CF-Data-Poll, and CF-Data-Ack-Poll. In the FIFO scheme, the PC transmits frames tostations according to the order of frames in its queue and piggybacks the polling frames to thecorresponding stations. If the PC does not receive ACK from the polled station, it keeps pollinguntil reaching the retry limit. If there are no frames in the PC queue, the FIFO scheme becomesRR scheme. When there is a new frame in the PC queue, the FIFO scheme is resumed. In thePriority scheme, the PC transmits traffic or poll frames to the stations from higher Type of Service(TOS) to lower TOS in turn. After all stations with TOS � 1 has been served, best-effort stationsare polled in round robin order until frames of TOS � 1 enter the PC queue. The PC alwayspiggybacks polling frame with data frame and piggybacks ACK frame if it needs to respond to

22


a station transmitting a data frame just now. For the last scheme, which is called Priority EffortLimited Fair (Priority-ELF), the PC transmits traffic or poll frames to the stations from higherTOS to lower TOS in turn. However, the PC checks counters that record the number of framestransmitted from each station in a given period. The PC only polls the stations with countervalues greater than 1, even if the PC has data for the station in its queue or the station has data tosend.

5.2 TDMA-like approaches

Ni et al. [34] mentions a method, in which TDMA-like time slots are set up and allocated tostations for service differentiation. Each station only transmits in its time slots and there is littlerequirement from the AP to intervene with packet transmissions.

Wei et al. [51] recommend a method called Slotted PCF (S-PCF). During CP, the PC identi-fies the stations and their priority in a list. The PC polls stations for transmission requests anddetermines the transmission order based on traffic priority. After the CP period, the PC sends abeacon frame with transmission parameters to all stations, assigning slots to stations from higherpriority to lower priority. The PC could terminate the CFP at any time with a Contention Free Ac-knowledgment (CFACK) frame. When a station receives the beacon frame, it sets its transmissionparameters such as NAV. If a station receives a traffic frame, it should reply with a small frame.A station always waits until its reserved slot to transmit a frame. If there are many transmissionrequests during S-PCF period, lower priority stations wait for the next DCF or S-PCF period. Astation sets new NAV after receiving a CFACK frame.

Qiang et al. [28, 38] introduce a method using distributed schedulers. At the AP, there is boththe Master Scheduler and one Slave Scheduler (SS) while there is only one SS at each station.The Master Scheduler assigns time slots to SSs and each SS then assigns its allocated time slotsamong its connections. A Contention Free Period Repetition Interval (CFPRI) consists of a periodfor contention-free traffic and a period for contention-based traffic. A maximum bandwidth por-tion (say 25%) is assgined for non-real-time traffic. If the number of Non-Real-Time Data Units(NRTDUs) reported to the Master Scheduler is less than or equal to 0.25 � the total number ofData Units (DUs) that could be supported in a CFPRI, then the unused bandwidth by NRTDUscould be allocated for Real-Time Data Units (RTDUs) and their delay could be decreased. TheNTRDUs have a large timeout value which is decremented with time at the SSs. When this valuereaches the timeout value of RTDUs, the NRTDU would be considered as a RTDU and its delayrequirement is reported to the Master Scheduler. If the number of NRTDUs is greater than 0.25� the total number of DUs that could be supported in a CFPRI, bandwidth portion for NRTDUswould be limited to 25%. Some NTRDUs staying long enough in the the queue would get somebandwidth. If a station has more time slots than the RTDUs (occur when the sum of total RTDUsand total NRTDUs reported to Master Scheduler is less than the total number of DUs supportedin a CFPRI), the unused slots could be used by the remaining stations or AP could end CFP andthe unused bandwidth is used in CP. If there is no NRTDUs, the total duration of CFPRI minustime for Association Request (AR) frame (used by a new station to associate with AP) transmis-sion is assigned for RTDUs.At the beginning of CFPRI, the Master Scheduler at AP allocates the bandwidth for each SS ac-cording to the data unit priorities. The assigned priority value depends on the remaining timeoutof DUs and channel error. When a station or the AP has data for transmission, the AssociationInformation (AI) is piggybacked in uplink data from remote SS to the Master Scheduler the in AP.An SS schedules for transmission of NRTDUs in CP. For those NRTDUs whose timeout reachesthe timeout value of RTDUs, SS sends this AI to the Master Scheduler. The bandwidth is sharedfairly between different DU categories and stations. After the beacon frame, the AP broadcastsPolling Information (PI) once, which includes the order of stations and their assigned time slotsfor transmission. If a station is allocated time slots in this CFP, the ACK and AI are piggybackedin its uplink data. If a station does not have time slots allocated, mini slots for transmission ofACK and AI would be allocated before the end of CFP. The AP broadcasts one ACK with a bitmapfor all successful transmission of uplink DUs in this CFP.

23


Zhao et al. [55] recommend an approach called Modified-PCF (M-PCF), in which stationsaccess the medium in a hub-poll manner. During CP, when a station successfully gains accessto the medium, the AP adds this station to its polling list, gives it a polling sequence numberand broadcasts to other stations. In CFP, after the first station is polled, all stations in pollinglist are allowed automatic access to the medium in turn without polling. Furthermore, a stationreceiving a non-real-time frame replies with an ACK, while a station receiving a real-time framedoes not.

5.3 Adaptive polling approach

Kim et al. [32] suggest a polling scheme for voice transmission. The Statistical Activity Detection(SAD) performs the following tasks in each phase of a voice activity. In phase 1, detection periodof talkspurt-to-silence state change, the SAD could detect the station changing to silence state if themore data field in the MAC header is set to false. During phase 2, silence period, the SAD assumesthere is no packet for transmission unless there is silence-to-talkspurt state change. Within phase 3,detection period of silence-to-talkspurt state change, after the arrival of a packet, the SAD assumes thevoice activity is changed to silence state and must stay there for a period of time at least Tthresholdbefore changing back to talkspurt state. In phase 4, talkspurt period, the SAD assumes the stationremains in talkspurt state as long as the more data field in the MAC header is set to true. Thereare four logical lists: pollable list, active node list, inactive node list and adaptive polling list.Each station is represented with an Association ID (AID). The pollable list has the AIDs of all thestations. The active and inactive lists keep the AIDs of stations currently in talkspurt and silencestates, respectively. The adaptive polling list is used by the AP to poll stations during CFP. Atthe beginning, all stations in the pollable list are assumed to be in talkspurt state. If a stationchanges to silence state, its AID is removed from the active and adaptive lists and is moved to theinactive list. The SAD has a timer for each inactive station, and when this timer exceeds Tthreshold,the station AID is moved from the inactive list to the adaptive list. If this station has a DATAframe to send, then its AID is added to the active list. The adaptive polling list is shorter than theexhaustive polling list in IEEE 802.11 round robin scheme.

5.4 Contention-based multipolling approach

Lo et al. [33] propose a contention-based Multipoll technique. This method reduces the number ofpolling frames that the PC sends, as compared with the SinglePoll method used in PCF or HCF, inwhich the PC sends one polling frame for each polled station in the polling list. Different backoffintervals are assigned to stations in the polling list and the stations perform backoff procedureafter receiving the CP-Multipoll frame. The polling order is transformed into the contendingorder, which is the same as the order of the assigned backoff interval in ascending order. In theCP-Multipoll, stations perform restricted backoff procedure without waiting for medium to beidle for DIFS, therefore contention of these stations is protected from other stations performingbackoff procedure in DCF mode. When a station receives a CP-Multipoll frame, this stationchecks whether it belongs to the polling list. If it is in the polling list, it sets its backoff time,maximum TXOP duration and determines the actual transmission duration. If the medium isidle, the backoff interval is decremented until this interval reaches zero, when RTS/CTS framesare exchanged and if there is no collision, DATA frame(s) are transmitted. If there is a collisionin RTS/CTS frame transmission, after three unsuccessful retransmission attempts, the stationaborts its transmission. If the medium is busy, the NAV is set after receiving a frame and thestation defers medium access until the time specified in the NAV. For the PC, if the medium isbusy, the PC performs the initial backoff after the medium is found idle for PIFS. The PC setsits internal backoff time and sends a CP-Multipoll frame. If it receives a RTS frame, it respondswith a CTS frame and also records the successful stations. When the medium is idle, the backoffinterval is decremented. When the backoff interval reaches zero, the PC sends a null CP-Multipollframe to cancel all the pending polled stations and polls the failed stations individually. A stationreceiving the null CP-Multipoll abandons the unfinished actions in the previous CP-Multipoll.

24


5.5 Discussion

Ganz’s scheme [22] works better than single poll scheme in noisy or hidden terminal conditionsbecause stations have more opportunities to receive the poll. However, each station has to lis-ten to the PC and other stations for the SuperPoll frame and the redundant polling frames usemore medium space. In Suzuki methods [47], for a given voice and video MPDU droppingprobability, priority schemes 1 and 2 can accommodate a larger number of multimedia stations.However, the non-priority scheme can offer higher data throughput. For scheme 2, the effect ofhigh-priority polling period is a trade-off between throughput of multimedia and data stations.Yeh’s priority and Priority-ELM schemes [54] can provide QoS and prevention against link errors.However, the priority scheme starves the best-effort traffic. The Priority-ELF scheme has highertotal throughput when there are link errors and also prevents starvation of best-effort traffic bet-ter than the priority scheme. For balanced traffic model, the total throughput of round robin andDCF schemes are similar and much higher than that of the FIFO scheme. On the other hand, withunbalanced traffic model, the total throughput of FIFO scheme is much higher than DCF scheme,which is higher than round robin scheme. In Ni’s and Wei’s methods [34, 51], stations only trans-mit in their allocated slots and there is no need for polling frames, thus more bandwidth can beused for traffic transmission. Mathematical analysis shows that Wei’s method performs betterthan DCF in terms of collision probability, bandwidth rate, average delay and data frame loss.Qiang’s approach [38] can guarantee the delay and throughput of real-time services and performsbetter than DCF in terms of throughput and delay. Zhao’s M-PCF [55] can support voice and datatraffic better than PCF with higher voice throughput and lower packet dropping probability inboth ideal channel or erroneous channel conditions. For Kim’s adaptive polling [32], in heavyload conditions, SAD and Hybrid Activity Detection (HAD) have good performance in terms ofgoodput and delay because unnecessary polling attempts are avoided while talkspurt occurrenceis detected. However, for video support, traffic characteristics of video need to be considered.

6 Summary and Conclusion

In this article, the operation of the legacy IEEE 802.11 MAC functions, including DCF and PCF,is described. The 802.11e enhancements to the original MAC, consisting of EDCF and HCF func-tions, is also introduced. Afterward, the QoS mechanisms for WLAN, based on the distributedand centralized approaches are summarized and evaluated. Generally, distributed methods aresimpler for implementation and have smaller overhead as compared with centralized ones. Dis-tributed protocols are also more flexible than centralized protocols when dealing with highlybursty traffic of real-time services. However, centralized mechanisms can guarantee bandwidthrequirements, while distributed mechanisms cannot. Distributed approaches can be further di-vided into priority-based and fair scheduling. In priority-based approach, the differentiationbased on IFS is simpler than mapping priority to CW or BI. Furthermore, it does not introducethroughput and delay fluctuation as does the approach based on CW or BI using a random BI.The priority-based mechanisms, however, do not give traffic equal opportunity for medium ac-cess and tend to starve lower priority traffic, hence the need for fair scheduling schemes.The fairscheduling methods usually employ additional fields in the MAC header and require stations lis-ten to every frame transmitted and inspect these fields to find out the load condition, the through-put and delay experienced by each flow. They also require the assignment of the flow weights,the calculation of the fairness index resulting in some complexity. Of all the methods evaluated,most of them only consider throughput guarantee but not the delay/jitter requirements. Theseaspects of QoS are of increasing importance for video streaming and anticipated increases to in-teractive video applications. Although the proposed techniques can offer QoS differentiation, inorder to guarantee QoS, admission control and resource allocation are also required. Moreover,in order to select the suitable QoS mechanism with the proper set of parameters, more researchincluding theoretical studies and simulations is required.

25


References

[1] IEEE Std. 802.11-1999. Part 11: Wireless LAN Medium Access Control (MAC) and PhysicalLayer (PHY) Specifications. ISO/IEC 8802-11:1999(E), 1999.

[2] IEEE Std. 802.11a 1999. Part 11: Wireless LAN Medium Access Control (MAC) and PhysicalLayer (PHY) Specifications: High-Speed Physical Layer in the 5 GHz Band, 1999.

[3] IEEE Std. 802.11b 1999. Part 11: Wireless LAN Medium Access Control (MAC) and PhysicalLayer (PHY) Specifications: Higher-Speed Physical Layer Extension in the 2.4 GHz Band,1999.

[4] IEEE 802.11e/D4.0. Draft Supplement to Part 11: Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) Specifications: Medium Access Control (MAC) Enhance-ments for Quality of Service (QoS), November 2002.

[5] IEEE Std. 802.1d 1998. Part 3: Media Access Control (MAC) Bridges, 1998.

[6] I. Aad and C. Castelluccia. Introducing Service Differentiation into IEEE 802.11. In FifthIEEE Symposium on Computers and Communications ISCC 2000, pages 438–443, July 2000.

[7] I. Aad and C. Castelluccia. Differentiation Mechanisms for IEEE 802.11. In Twentieth AnnualJoint Conference of the IEEE Computer and Communications Societies INFOCOM 2001 [23], pages209–218.

[8] I. Aad and C. Castelluccia. Remarks on Per-flow Differentiation in IEEE 802.11. In EuropeanWireless EW 2002, Florance, Italy, February 2002.

[9] A. Banchs and X. Perez. Distributed Weighted Fair Queuing in 802.11 Wireless LAN. InIEEE International Conference on Communications ICC 2002 [24], pages 3121–3127.

[10] A. Banchs and X. Perez. Providing Throughput Guarantees in IEEE 802.11 Wireless LAN.In IEEE Wireless Communications and Networking Conference WCNC 2002, volume 1, pages130–138. IEEE Communications Society, March 2002.

[11] A. Banchs, X. Perez, M. Radimirsch, and H. J. Stuttgen. Service Differentiation Extensions forElastic and Real-time Traffic in 802.11 Wireless LAN. In IEEE Workshop on High PerformanceSwitching and Routing, pages 245–249, May 2001.

[12] M. Barry, A. T. Campbell, and A. Veres. Distributed Control Algorithms for Service Dif-ferentiation in Wireless Packet Networks. In Twentieth Annual Joint Conference of the IEEEComputer and Communications Societies INFOCOM 2001 [23], pages 582–590.

[13] M. Benveniste. Overview of Tier Contention Multiple Access (TCMA). IEEE Document,802.11-01/019, January 2001.

[14] M. Benveniste. Proposed Normative Text for TCMA with Backoff Adaptation. IEEE Docu-ment, 802.11-01/135r1, March 2001.

[15] M. Benveniste. Tiered Contention Multiple Access (TCMA), a QoS-based Distributed MACProtocol. In The 13th IEEE International Symposium on Personal, Indoor and Mobile Radio Com-munications, volume 2, pages 598–604, September 2002.

[16] Wen-Tsuen Chen, Bo-Bin Jian, and Shou-Chih Lo. An Adaptive Retransmission Schemewith QoS Support for the IEEE 802.11 MAC Enhancement. In IEEE 55th Vehicular TechnologyConference VTC 2002 [26], pages 70–74.

[17] G. Chesson, A. Singla, W. Diepstraten, and H. Tenniusen et. al. VDCF State Machine De-scription. IEEE Document, 802.11-00/412r1, November 2000.

26


[18] G. Chesson, A. Singla, W. Diepstraten, and H. Tenniusen et. al. VDCF Proposed Draft Text.IEEE Document, 802.11-01/131, March 2001.

[19] J. Deng and R. S. Chang. A Priority Scheme for IEEE 802.11 DCF Access Method. IEICETransactions on Communications, E82-B(1):96–102, January 1999.

[20] A. Dugar, N. Vaidya, and P. Bahl. Priority and Fair Scheduling in a Wireless LAN. InIEEE Military Communications Conference MILCOM 2001, Communications for Network-CentricOperations: Creating the Information Force, volume 2, pages 993–997, October 2001.

[21] Zuyuan Fang, Brahim Bensaou, and Yu Wang. Performance Evaluation of a Fair BackoffAlgorithm for IEEE 802.11 DFWMAC. In 3rd ACM International Symposium on Mobile AdhocNetworking and Computing, pages 48–57, Lausanne, Switzerland, July 2002.

[22] A. Ganz, A. Phonphoem, and Z. Ganz. Robust Superpoll with Chaining Protocol for IEEE802.11 Wireless LANs in Support of Multimedia Applications. Wireless Networks, 7(1):65–73,2001.

[23] IEEE Communications Society. Twentieth Annual Joint Conference of the IEEE Computer andCommunications Societies INFOCOM 2001, April 2001.

[24] IEEE Communications Society. IEEE International Conference on Communications ICC 2002,April 2002.

[25] IEEE Communications Society. IEEE Wireless Communications and Networking ConferenceWCNC 2003, March 2003.

[26] IEEE Vehicular Technology Society. IEEE 55th Vehicular Technology Conference VTC 2002, May2002.

[27] IEEE Vehicular Technology Society. The 57th IEEE Semiannual Vehicular Technology ConferenceVTC 2003, April 2003.

[28] L. Jacob, R. Radhakrishna Pillai, and B. Prabhakaran. MAC Protocol Enhancements and aDistributed Scheduler for QoS Guarantees over the IEEE 802.11 Wireless LANs. In IEEE56th Vehicular Technology Conference VTC 2002, volume 4, pages 2410–2413. IEEE VehicularTechnology Society, September 2002.

[29] Kanghee Kim, A. Ahmad, and Kiseon Kim. A Wireless Multimedia LAN Architecture Us-ing DCF with Shortened Contention Window for QoS Provisioning. IEEE CommunicationsLetters, 7(2):97–99, February 2003.

[30] Kanghee Kim, A. Ahmad, and Kiseon Kim. A Wireless Multimedia LAN Architecture UsingDCF with Shortened Contention Window for QoS Provisioning. In IEEE Wireless Communi-cations and Networking Conference WCNC 2003 [25], pages 1308–1311.

[31] Kanghee Kim, Seokjoo Shin, and Kiseon Kim. A Novel MAC Scheme for Prioritized Servicesin IEEE 802.11a Wireless LAN. In Joint 4th IEEE International Conference on ATM (ICATM2001) and High Speed Intelligent Internet Symposium, pages 196–199, April 2001.

[32] Young-Jae Kim and Young-Joo Suh. Adaptive Polling MAC Schemes for IEEE 802.11 Wire-less LANs. In The 57th IEEE Semiannual Vehicular Technology Conference VTC 2003 [27], pages2528–2532.

[33] Shou-Chih Lo, Guanling Lee, and Wen-Tsuen Chen. An Efficient Multipolling Mechanismfor IEEE 802.11 Wireless LANs. IEEE Transactions on Computers, 52(6):764–778, June 2003.

[34] Q. Ni, L. Romdhani, T. Turletti, and I. Aad. QoS Issues and Enhancements for IEEE 802.11Wireless LAN. Technical Report 4612, INRIA, November 2002.

27


[35] M. Ogawa, I. Shimojima, and T. Hattori. CoS Guarantee Control for Wireless LAN. In IEEE55th Vehicular Technology Conference VTC 2002 [26], pages 50–54.

[36] W. Pattara-Atikom, P. Krishnamurthy, and S. Banerjee. Distributed Mechanisms for Qualityof Service in Wireless LANs. IEEE Wireless Communications, 10(3):26–34, June 2003.

[37] W. Pattara-Aukom, S. Banerjee, and P. Krishnamurthy. Starvation Prevention and Quality ofService in Wireless LANs. In The 5th International Symposium on Wireless Personal MultimediaCommunications, volume 3, pages 1078–1082, October 2002.

[38] Qiu Qiang, L. Jacob, R. Radhakrishna Pillai, and B. Prabhakaran. MAC Protocol Enhance-ments for QoS Guarantee and Fairness over the IEEE 802.11 Wireless LANs. In EleventhInternational Conference on Computer Communications and Networks, pages 628–633, October2002.

[39] Daji Qiao and K. G. Shin. Achieving Efficient Channel Utilization and Weighted Fairnessfor Data Communications in IEEE 802.11 WLAN under the DCF. In Tenth IEEE InternationalWorkshop on Quality of Service, pages 227–236, May 2002.

[40] L. Romdhani, Qiang Ni, and T. Turletti. Adaptive EDCF: Enhanced Service Differentiationfor IEEE 802.11 Wireless Ad-hoc Networks. In IEEE Wireless Communications and NetworkingConference WCNC 2003 [25], pages 1373–1378.

[41] Shiann-Tsong Sheu and Tzu-Fang Sheu. DBASE: a Distributed Bandwidth Alloca-tion/Sharing/Extension Protocol for Multimedia over IEEE 802.11 Ad hoc Wireless LAN.In Twentieth Annual Joint Conference of the IEEE Computer and Communications Societies INFO-COM 2001 [23], pages 1558–1567.

[42] Shiann-Tsong Sheu, Tzu-Fang Sheu, Chih-Chiang Wu, and Jiau-Yu Luo. Design and Imple-mentation of a Reservation-Based MAC Protocol for Voice/Data over IEEE 802.11 Ad-hocWireless Networks. In IEEE International Conference on Communications ICC 2001, volume 6,pages 1935–1939. IEEE Communications Society, June 2001.

[43] J. Sobrinho and A. S. Krishnakumar. Real-time Traffic over the IEEE 802.11 Medium AccessControl Layer. Bell Labs Technical Journal, 1(2):172–187, 1996.

[44] J. L. Sobrinho and A. S. Krishnakumar. Distributed Multiple Access Procedures to ProvideVoice Communications over IEEE 802.11 Wireless Networks. In Global TelecommunicationsConference GLOBECOM ’96, volume 3, pages 1689–1694, November 1996.

[45] J. L. Sobrinho and A. S. Krishnakumar. Quality-of-Service in Ad hoc Carrier Sense MultipleAccess Wireless Networks. IEEE Journal on Selected Areas in Communications, 17(8):1353–1368,August 1999.

[46] Nah-Oak Song, Byung-Jae Kwak, Jabin Song, and Leonard E. Miller. Enhancement of IEEE802.11 Distributed Coordination Function with Exponential Increase Exponential DecreaseBackoff Algorithm. In The 57th IEEE Semiannual Vehicular Technology Conference VTC 2003[27], pages 2775–2778.

[47] T. Suzuki and S. Tasaka. Performance Evaluation of Priority-Based Multimedia Transmis-sion with the PCF in an IEEE 802.11 Standard Wireless LAN. In The 12th IEEE InternationalSymposium on Personal, Indoor and Mobile Radio Communications, volume 2, pages G70–G77,September 2001.

[48] N. H. Vaidya, P. Bahl, and S. Gupta. Distributed Fair Scheduling in a Wireless LAN. InThe 6th Annual International Conference on Mobile computing and Networking, pages 167–178,Boston, Massachusetts, United States, 2000.

28


[49] A. Veres, A. T. Campbell, M. Barry, and Li-Hsiang Sun. Supporting Service Differentiationin Wireless Packet Networks Using Distributed Control. IEEE Journal on Selected Areas inCommunications, 19(10):2081–2093, October 2001.

[50] Y. Wang and B. Bensaou. Achieving Fairness in IEEE 802.11 DFWMAC with Variable PacketLengths. In Global Telecommunications Conference GLOBECOM ’01, volume 6, pages 3588–3593, November 2001.

[51] A. Wei and S. Boumerdassi. A New PCF Scheme for Multimedia Traffic in IEEE 802.11Wireless LAN. In IASTED International Conference Computer Science and Technology CST 2003,May 2003.

[52] G. W. Wong and R. W. Donaldson. Improving the QoS Performance of EDCF in IEEE 802.11eWireless LANs. In IEEE Pacific Rim Conference on Communications, Computers and Signal Pro-cessing PACRIM 2003, volume 1, pages 392–396, August 2003.

[53] Jin Xiaohui, Li Jiandong, and Guo Feng. M-DCF: a MAC Protocol Supporting QoS in Ad hocNetwork. In International Conference on Communication Technology WCC-ICCT 2000, volume 2,pages 1718–1721, August 2000.

[54] Jing-Yuan Yeh and Chienhua Chen. Support of Multimedia Services with the IEEE 802-11MAC Protocol. In IEEE International Conference on Communications ICC 2002 [24], pages 600–604.

[55] Liqiang Zhao and Changxin Fan. M-PCF: Modified IEEE 802.11 PCF Protocol ImplementingQoS. Electronics Letters, 38(24):1611–1613, November 2002.

29


Documents

Department of Electrical and Computer Systems …Wireless Local Area Network), ... CoS Class of Service CP Contention Period 2 MECSE-5-2004: ... NAV Network Allocation Vector